WTR Small-Cap Spotlight

James Kisner analyzes Anthropic IPO valuation, profitability, and risks using EV/S multiples and compute cost trends.

11 min Listen to the original
IPO ValuationAI InfrastructureCompute CostsVenture CapitalEnterprise SoftwareTech IPOs

Summary

James Kisner evaluates Anthropic's upcoming IPO, arguing it appears cheaper than peers on an EV/S basis but faces significant revenue concentration and pricing pressure risks. He highlights that while Anthropic is profitable with a 15% adjusted operating margin, its compute costs surged 77% sequentially, outpacing revenue growth. Kisner notes that over 80% of Q3 revenue comes from models released in the last six months, indicating high churn risk. He also discusses the dependency of Anthropic's customer base on venture capital and the disciplined financing stance of backer NVIDIA.

Key takeaways

  1. 1
    the enterprise value is about $1.1 trillion. That's 32 times trailing sales, well below the 53 times average for the nine technology IPOs in our note. Cerebris and SpaceX both priced above 90 times.

    Unidentified speaker (speaker_0) · 1:15

  2. 2
    Fable 5.2 launched with token prices 20% below its predecessors, so Anthropic needs 25% more comparable volume in tokens just to hold revenue flat.

    Unidentified speaker (speaker_0) · 5:39

Episode sections

  1. 0:00 – 0:50

    Introduction and Methodology

    James Kisner introduces the topic of the Anthropic IPO and outlines the top 10 considerations for investors. He clarifies that the analysis uses Enterprise Value to Sales (EV/S) rather than Market Cap to Sales to account for cash and debt.

  2. 0:50 – 3:47

    Valuation and Profitability Metrics

    Kisner argues Anthropic appears cheaper than other tech IPOs on an EV/S basis, citing a 32x trailing multiple. He notes the company is already generating adjusted operating profit with a 15% margin, contrasting this with negative margins at Cerebris, SpaceX, and early Facebook.

  3. 3:47 – 5:39

    Revenue Concentration and Compute Costs

    Over 80% of Q3 revenue comes from models released within the last six months, implying a short shelf life. Compute spending surged 77% sequentially to $9.9 billion, with training costs being a significant, undisclosed portion of this expense.

  4. 5:39 – 7:20

    Pricing Pressure and Customer Base

    Anthropic faces pricing pressure, with newer models launching at lower token prices and blended closed model prices falling 57% industry-wide. While enterprise adoption is broadening, a significant portion of customers are startups dependent on venture capital funding.

  5. 7:20 – 8:20

    Investor Economics and Geopolitics

    Backers Google and Amazon face margin pressure and negative free cash flow, potentially making them more demanding partners. Tighter China chip restrictions may hinder rival Chinese models, offering marginal benefit to Anthropic.

  6. 8:20 – 11:11

    NVIDIA Leadership and IPO Risks

    New interim CEO Colette Kress signals disciplined, selective financing support from NVIDIA, contrasting with previous aggressive tactics. Kisner concludes that successful IPOs do not guarantee investment success, noting only 4 of 9 recent tech IPOs beat the Nasdaq benchmark.

Claims worth verifying

Anthropic's enterprise value is about $1.1 trillion at the $74 midpoint, resulting in a 32x trailing EV/S multiple.

Unidentified speaker (speaker_0) ·1:15

How to verify: Check Anthropic's S-1 filing for cash, debt, and shares outstanding; calculate EV and compare to reported sales figures. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Anthropic's Q3 revenue grew 47% sequentially to $17 billion with a 15% adjusted operating margin.

Unidentified speaker (speaker_0) ·2:57

How to verify: Check Anthropic's Q3 earnings release or S-1 filing for revenue and operating income data. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

More than 80% of Q3 revenue came from models released within the prior six months.

Unidentified speaker (UNKNOWN) ·3:50

How to verify: Check Anthropic's S-1 filing for revenue breakdown by product/model vintage. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Anthropic spent $9.9 billion on computing through Q3, up 77% sequentially.

Unidentified speaker (speaker_0) ·4:10

How to verify: Review Anthropic's S-1 filing for cash flow from operations and capital expenditures related to compute. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Fable 5.2 launched with token prices 20% below its predecessors.

Unidentified speaker (speaker_0) ·5:39

How to verify: Check Anthropic's product announcement or pricing page for Fable 5.2 vs. previous models. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

AI companies took more than 80% of US venture dollars in the first nine months of 2026.

Unidentified speaker (speaker_0) ·6:53

How to verify: Check PitchBook and NVCA reports for Q1-Q3 2026 venture capital deployment data. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Only four of nine technology IPOs beat the Nasdaq total return benchmark over their holding periods through November 13th.

Unidentified speaker (UNKNOWN) ·9:48

How to verify: Calculate total returns for the nine IPOs mentioned in the report against Nasdaq data from the offering date to November 13, 2026. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Do your own research

  • Verify Anthropic's S-1 filing for detailed revenue breakdown by model release date and customer concentration.
  • Check NVIDIA's latest 10-K or earnings call transcripts for updates on 'disciplined' financing commitments to Anthropic.
  • Review industry token pricing trends from major AI providers to confirm the 57% industry-wide price drop claim.
  • Analyze Anthropic's compute spend vs. revenue growth in subsequent quarterly filings to assess training cost scalability.
  • Compare Anthropic's EV/S multiple against the S-1 filing's stated valuation and recent comparable tech IPOs.
  • Assess the dependency of Anthropic's startup customer base on VC funding cycles by reviewing broader VC investment data.
  • Look for counter-arguments regarding the shelf life of AI models and potential long-term contract renewals.
  • Verify the 15% adjusted operating margin calculation by reviewing the specific non-cash charges excluded in the filing.
Full transcriptexpand
0:00
Unidentified speaker (speaker_0): Welcome to another WTR Analyst Perspectives Podcast.
0:05
Unidentified speaker (UNKNOWN): In
0:05
Unidentified speaker (speaker_0): this series, we step back from our usual focus on small cap companies and dedicate the space to our analysts' own read on where a sector stands and what investors should be watching.
0:15
Unidentified speaker (UNKNOWN): I'm
0:16
Unidentified speaker (speaker_0): James Kisner, Managing Director and Analyst, covering the technology sector with a focus on artificial intelligence at Watertower Research.
0:23
Unidentified speaker (UNKNOWN): Today
0:23
Unidentified speaker (speaker_0): I want to discuss a hot topic for investors at the moment: the Enthropic IPA. IPO.
0:27
Unidentified speaker (UNKNOWN): I
0:28
Unidentified speaker (speaker_0): encourage you to head on over to WatertowerResearch.com and check out my report we published on Monday on the topic, which we titled Enthropic IPO: The Greatest Show in Tech. All it takes is an email address, and these insights are all yours to enjoy free of charge.
0:41
Unidentified speaker (UNKNOWN): To
0:42
Unidentified speaker (speaker_0): make it a little more fun, I made a top 10 list.
0:45
Unidentified speaker (UNKNOWN): Here
0:45
Unidentified speaker (speaker_0): are the top 10 things we think investors should consider with respect to the anthropic IPO.
0:50
Unidentified speaker (UNKNOWN): Number
0:50
Unidentified speaker (speaker_0): one, in the Enthropic looks less expensive than several high-profile tech IPOs.
0:55
Unidentified speaker (UNKNOWN): You know,
0:55
Unidentified speaker (speaker_0): a quick note-on method, um, we're not using mark cap of sales, uh, which we think is a bit of a flawed metric. You know, we use enterprise value sales, which accounts for cash and debt. Um, you know, so with that clarification, at the original $74 midpoint, and encountering counting the private placement with Google, NVIDIA and friends.
1:14
Unidentified speaker (UNKNOWN): Um
1:15
Unidentified speaker (speaker_0): the enterprise value is about $1.1 trillion. That's 32 times trailing sales, well below the 53 times average for the nine technology IPOs in our note. Cerebris and SpaceX both priced above 90 times. Now, let's take a look at forward multiples, which I think are potentially more important to investors, tend to care more about the future than the
1:34
Unidentified speaker (speaker_0): past. Um In our note, we use an illustrative case where revenue grows $150. 159% next year and 62% in 2028, which is a big deceleration from over a thousand percent growth in Q3, but on a larger base.
1:48
Unidentified speaker (speaker_0): In terms of revenue, those growth rates tied a 2026, 2027, and 2028 revenue of 56 billion, a little
1:57
Unidentified speaker (speaker_0): 150 billion, and uh a little under 240 billion respectively. So uh again, at the $74 midpoint, it works out to about eight times uh forward uh 20 to 27 sales and five times 2028. Um that's pretty close to our US pure medians of 8.0 and 5.1.
2:17
Unidentified speaker (speaker_0): Um it does look like the price could go higher. The Wall Street Journal reports pricing could exceed uh $80, $86. Uh again, just an illustration. it's 9.1 times 2027 on an EV to sales basis and 5.6 times on a 28 to 28 EV to sales basis, you know,
2:37
Unidentified speaker (speaker_0): a bit above peers, but not exactly nose bleed valuations. Of course, higher pricing means higher expectations and more risk if there's a hiccup in the business. Number two, Anthropic is already pretty profitable for a company. this young and fast growing. Cuthi revenue grew 47% sequentially to $17 billion,
2:57
Unidentified speaker (speaker_0): and they delivered a 15% adjusted operating margin, which excludes things like non-cash, shitable charges. So
3:05
Unidentified speaker (speaker_0): with those caveats aside,
3:08
Unidentified speaker (speaker_0): gap operating margin was negative 3%, still within spitting distance of break-even. Um, you know, With a caveat that adjustments aren't identical. Cerebrus had a negative 10% operating margin uh before they went public. Consolidated SpaceX, including its AI business, um, had a negative 28%. You know, Facebook was the standout uh at a positive 46% um
3:27
Unidentified speaker (speaker_0): adjusted operating margin. Um, but nonetheless, um, you know, researchable profitability here, uh, Anthropic and Q3. Um, what's notable is that Anthropic has recently started generating uh real adjusted operating profit growing at an extraordinary rate. And where that lasts, you know, takes us to the next point. Number three, most of Anthropic's revenue
3:47
Unidentified speaker (speaker_0): comes from models less than six years old.
3:50
Unidentified speaker (UNKNOWN): More
3:50
Unidentified speaker (speaker_0): than 80% of Q3 revenue came from the models released within the prior six months. Opus 5.5 arrived just 60 days after Opus V. Customers moved move to new models quickly. So this revenue arguably has a short. shelf life. To keep it, Anthropic has to keep shipping better models, which makes model development a cost of staying in business from our
4:10
Unidentified speaker (speaker_0): perspective. So, you know, related to that, let's talk about compute. Compute spending is going up sharply, and training costs are a major and unknown. Enthropic spent 9.9 billion on computing through Q3 three. That was up 77% sequentially. You know, that's that's faster uh than than. th revenue. Uh commitments due over the past 12 months through September
4:30
Unidentified speaker (speaker_0): 27 average uh 17.7 billion a quarter, 79% above that.
4:35
Unidentified speaker (UNKNOWN): Uh
4:35
Unidentified speaker (speaker_0): payments and expenses won't line up exactly, and in some new capacity replaces existing spending. But you know, I think the direction's pretty clear here. Uh we should also talk about training costs. So that 9.9 billion, at least 7.4 billion of that Q3 compute set outside. cost of revenue, including a research and model development. And that's conservative
4:55
Unidentified speaker (speaker_0): estimates. It's cost of revenue also includes some personnel and overhead.
4:59
Unidentified speaker (UNKNOWN): So
5:00
Unidentified speaker (speaker_0): training alone isn't disclosed, but we think it's a large chunk of that $7.4 billion
5:07
Unidentified speaker (speaker_0): of that compute is for model training. What's not known is how much that training cost could go up with ever larger and more performant models. Epoch AI estimates frontier training compute has grown about five times a year, far faster than chip performance per dollar. On the other side, algorithms keep getting more efficient. We don't know how those forces
5:26
Unidentified speaker (speaker_0): balance, and it matters. Every $5 billion of annual training expense moves the operating margin by more than three points on our 2027 revenue case. Number five, pricing pressure is already material.
5:39
Unidentified speaker (speaker_0): Fable 5.2 launched with token prices 20% below its predecessors, so Anthropic needs 25% more comparable volume in tokens just to hold revenue flat. Anthropic itself disclosed roughly 600 million of sequential revenue pressure from price cuts and discounts on Q3. And Google is competing hard. Artificial analysis intelligence
5:59
Unidentified speaker (speaker_0): index, though Fable 5.2 still leads.
6:03
Unidentified speaker (speaker_0): Of course, a single score isn't a perfect measure, but it tells you the gap at the top appears somewhat narrow. Carbon's promotional token prices are half of Opus's through November 30th.
6:13
Unidentified speaker (UNKNOWN): More
6:13
Unidentified speaker (speaker_0): broadly, OpenRouter
6:16
Unidentified speaker (speaker_0): disclosed that their blended closed model prices fell 57% from January through October.
6:20
Unidentified speaker (UNKNOWN): The
6:21
Unidentified speaker (speaker_0): question is whether volume keeps out running price declines. Number six, enterprise adoption gives the gross growth story substance. Customers generated at least $100,000 a year,
6:33
Unidentified speaker (UNKNOWN): uh,
6:33
Unidentified speaker (speaker_0): went from more than $950 to more than $11,900 in a year. Um, the top five customers' share of quarterly revenue fell from 36% in Q1 to 21% in Q3, you know, which suggests uh you know that the growth is broader. Um, but it's also worth asking who's paying here. I mean, throw Throughout's up a own filing says a significant portion of its customers
6:53
Unidentified speaker (speaker_0): are startups and technology-oriented enterprises, and warns that some depend on venture capital and may cut spending if that funding dries
7:00
Unidentified speaker (UNKNOWN): up.
7:01
Unidentified speaker (speaker_0): AI companies took more than 80% of US venture dollars in the first nine months of 2026, according to PitchBook and NVCA, though that includes the lab's own giant rounds. Spending funded by venture rounds lasts only as long as the rounds. keep coming. Number seven, Anthropic's biggest backers face pressure on their own
7:20
Unidentified speaker (speaker_0): economics. Google and Amazon invest in Anthropic and supply its compute. They also distribute claw. Google acknowledged that AI mode monetizes below traditional research, and Alphabet lost roughly $500 billion in market value after earnings. Amazon's free cash flow was negative $33.5 billion last quarter as the their infrastructure spending
7:40
Unidentified speaker (speaker_0): surged.
7:41
Unidentified speaker (UNKNOWN): These
7:41
Unidentified speaker (speaker_0): companies can still fund AI, but pressure on their own returns could make them more demanding partners. Number eight, tighter China restrictions could help anthropic at the margin.
7:52
Unidentified speaker (UNKNOWN): New
7:53
Unidentified speaker (speaker_0): rules target Chinese access to advanced American ships through overseas data centers. If that's effective, they could make rival models harder hard to train. Again, we're talking about like Moonshot and Deep Sheet. uh DeepSeek.
8:05
Unidentified speaker (UNKNOWN): Um,
8:05
Unidentified speaker (speaker_0): Anthropic argues that tighter chip access also limits illicit distillation where customers train on its model's answers. So enforceable determine how much protection these rules provide, but ultimately seems like uh could be at the margin good news for anthropic.
8:20
Unidentified speaker (UNKNOWN): Uh
8:20
Unidentified speaker (speaker_0): number nine, there's a new sheriff at NVIDIA, which could be a mixed bag for Anthropic in our view.
8:26
Unidentified speaker (UNKNOWN): Um
8:27
Unidentified speaker (speaker_0): her name is Colette Cress, NVIDIA's interim CEO, um former CFO.
8:31
Unidentified speaker (UNKNOWN): And
8:31
Unidentified speaker (speaker_0): Jensen had swagger and a magic leather jacket. You know,
8:35
Unidentified speaker (UNKNOWN): I
8:35
Unidentified speaker (speaker_0): struggle to see NVIDIA getting more aggressive about financing customers from here. Um
8:41
Unidentified speaker (UNKNOWN): you know,
8:42
Unidentified speaker (speaker_0): on the November seventeenth call,
8:44
Unidentified speaker (UNKNOWN): uh
8:44
Unidentified speaker (speaker_0): Crest said that financing finance support would remain disciplined and selective, and if data center projects are delayed, our guarantee doesn't. exist anymore. Um
8:52
Unidentified speaker (UNKNOWN): she
8:53
Unidentified speaker (speaker_0): also did repeat last quarter's view that labs NVIDIA supports with its balance sheet would be about a quarter of next year's business, though she raised the overall growth outlook. So um
9:02
Unidentified speaker (UNKNOWN): you know
9:03
Unidentified speaker (speaker_0): I think the Bank of AI remains open largely. Um my concern is you know how to interpret uh you know uh discipline selective like how selective future support becomes
9:16
Unidentified speaker (speaker_0): Anthropics IPO cache helps, but its expansion still depends on suppliers getting facilities financed and built. There's a potential upside here too. The turmoil could push some labs towards AMD or Cerberus Hardware and custom chips developed with Broadcom. And that could improve uh the bargaining power uh of these labs
9:36
Unidentified speaker (speaker_0): over time, including Anthropic.
9:38
Unidentified speaker (UNKNOWN): Number
9:38
Unidentified speaker (speaker_0): 10. just you know, always worth noting, a successful IPO doesn't guarantee a successful investment.
9:44
Unidentified speaker (UNKNOWN): Um,
9:44
Unidentified speaker (speaker_0): of the nine technology IPOs we looked at, only four uh
9:48
Unidentified speaker (UNKNOWN): you know
9:48
Unidentified speaker (speaker_0): beat the Nasdaq total return benchmark over their respective holding periods through November 13th. Um, you know, those returns start at the offering price also, and not everybody has access to that price. Buying after what's often a first-day first day surge uh historically, um, what also? would also raise the hurdle.
10:05
Unidentified speaker (speaker_0): So keeping the original $74 midpoint in our illustrative $86 scenario separate
10:11
Unidentified speaker (speaker_0): from the price on your screen, you know, what return
10:13
Unidentified speaker (UNKNOWN): can
10:14
Unidentified speaker (speaker_0): Entropic deliver at the price you're actually paying? A credit order book
10:20
Unidentified speaker (speaker_0): won't answer that necessarily. So there you have it. The top 10 ten things we think investors should consider with respect to the Entropicity. Anthropic IPO. Again, be sure to check out my full report on Anthropic Published Monday, as well as visit www.watertowerresearch.com. You'll find my latest research on technology and AI,
10:40
Unidentified speaker (speaker_0): including my new AI newsletter, which I pour a lot of myself through these days, along with countless other small cap equity research reports, podcasts, fireside chats, and conference schedules from my myself and other analysts covering some of the most interesting sectors, including sustainable energy, healthcare, chemicals, natural resources, industrial technology, all from our experienced stable
11:00
Unidentified speaker (speaker_0): analyst at Water Tower Research. We will see you next time for another edition of the WTR Analyst Perspectives Podcast.
This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights